This approach violates alignment requirements causing misaligned accesses that on some architectures is mitigated by generating extra instructions while on some

This approach violates alignment requirements causing misaligned accesses that on some architectures is mitigated by generating extra instructions while on some other architectures it's a violation (e.g. segfault). There is no way for a compiler to infer the alignment requirement of a struct because it does not see its definition. You would always have to align the char buffer by hand but you cannot do it for the same reason compiler cannot. What you can do is to always greedily align the char buffer to the strictest (largest) fundamental requirement for that platform - in other words alignas(max_align_t). And this is actually what alloca() does for you by default. From https://gcc.gnu.org/onlinedocs/gcc/Other-Builtins.html https://gcc.gnu.org/onlinedocs/gcc/Other-Builtins.html > The object is aligned on the default stack alignment boundary for the target determined by the BIGGEST_ALIGNMENT_macro.